(Photo: 123RF)
Artificial intelligence (AI) has been talked about a lot in recent years, especially since the arrival of conversational robots, such as ChatGPT and Bard, launched respectively in November 2022 and March 2023. Behind this powerful software lies an infrastructure that will require investments of more than $1 trillion (US$1 billion) over the next decade, which may be of more interest to investors than AI itself.
This is also the opinion of the financial analysis company Huatai Financial Holdings, of Hong Kong, which revealed last July that it preferred, in order, companies present in the infrastructure sector computing, semiconductors and other components, conversational robots and applications.
The note, which focuses on the effects of the emergence of large language models, cites data from OpenAI, the company behind ChatGPT, which indicates that the computing power allocated to AI around the world doubles every three years. at four months. This rate is much faster than Moore’s Law, which suggests that computer power should double every 18 to 24 months.
At the end of August, analyst Daniel Ives of the American company Wedbush estimated that over the next ten years, cloud computing giants would spend a whopping US$1,000 billion on graphics processors. , servers and other EXPERT ADVICE Bank on suppliers to avoid fluctuations linked to technological innovations components to increase the capacity of data centers to feed the ever-growing appetite for analysis necessary for training and AI improvement.
“The main problem for companies that make the graphics processors that train large language models (LLMs) in AI is that they are not able to meet the demand at this time. moment,” say experts from the Fundamental Equity Management team at Manulife Investment Management. They argue that Nvidia (NVDA, US$460.18) owns the vast majority of the market for graphics processors used in data centers dedicated to artificial intelligence at the moment, but that the competition is organizing to get its share a piece of cake in this lucrative market. The company markets, among other things, its popular A100 and H100 models, which sell for several tens of thousands of US dollars each.
“We should also remember that Nvidia sources its highest-performance chips from Taiwan Semiconductors Manufacturing Company (TSM, US$93.10), better known by the acronym TSMC,” they say. TSMC’s stock trades at a ratio of just over 16 times its trailing 12-month earnings per share, according to Yahoo Finance, which is much more reasonable than Nvidia’s (111 times). Manulife experts attribute this weakness to the geopolitical situation in Taiwan.
Again, we must not forget that AI is still an emerging industry and only represents a small portion of the revenues of companies present in the sector. For example, analyst Sunny Lin of UBS estimates that the portion of TSMC’s revenues linked to AI should range between 3% and 4.6% in 2024 and between 8% and 11% in 2025.
Who will win the race?
That’s the multi-billion dollar question. For the moment, some experts believe it is better to avoid responding by investing in the suppliers of all the current leaders in the sector.
Bouthillier Capital (B-Cap) president and portfolio manager Mathieu Bouthillier puts companies like Broadcom (AVGO, US$851.82) on par with TSMC. “On the other hand, Applied Materials (AMAT, US$144.36), which we hold in our portfolio, produces the components that allow them to produce wafers and semiconductors,” he says.
For his part, John Plassard, investment specialist at Mirabaud, also cites the Dutch ASML Holding (ASML, US$651.01) as a supplier for TSMC, Broadcom and Marvell Technology (MRVL, US$53.50). He does not hesitate to describe this sector as “the hidden face of artificial intelligence”.
“With an indirect investment, you become a little more agnostic about who will win the artificial intelligence race, because the company that will have the best technology will need equipment,” explains Mathieu Bouthillier.
Cloud computing giants are getting organized
And data center operators are going to need equipment. A McKinsey & Company study estimates that global spending on data center construction will reach US$31 billion in 2022 and predicts that it will grow at a compound annual rate of 5.4% per year by 2030 for reach US$49 billion.
“The companies at the forefront of the data center industry form an oligopoly between Amazon Web Services (AWS), a subsidiary of Amazon (AMZN, US$133.26), Microsoft Azure (MSFT, US$322.98) ) and Google Cloud, a subsidiary of Alphabet (GOOGL, US$129.88),” maintains Mathieu Bouthillier.
These companies, like all other data center operators, have cost reduction goals and are working to design custom-made processors for their infrastructures. To achieve this, they need software specialized in digital design. “On this side, two companies, Cadence Design Systems (CDNS, US$232.51) and Synopsys (SNPS, US$442.24), control 70% of the global market,” says the B-Cap manager.
The portfolio manager says he monitors these two companies carefully, while specifying that the securities, although of “good quality”, are expensively valued. “But on the stock market, if you only buy inexpensive stocks, you often end up with a portfolio that is not worth much,” he says, adding that even in the event that the economic environment deteriorates almost everywhere in the world, around the world, developers will continue to work on their next iterations of microprocessors, which means that both software designers could do well.
We must not forget Advanced Micro Devices (AMD, US$102.25), which announced last June the launch of a new processor intended for AI during the third quarter, which should be followed by the start of mass production in the fourth quarter.
Focus on cooling systems?
John Plassard pushes the thinking even further by affirming that the construction of this computing power has a significant environmental cost. Citing the French magazine Le Big Data, the latter argues that data centers produce 2% of greenhouse gas emissions worldwide, which is comparable to the carbon footprint of the airline industry. This proportion could rise to 14% by 2040, according to the same publication.
“The more big data we have, the more data centers we have, the more there is a need to cool them. In some countries, data centers are even installed underwater because it is cooler there, rather than using air conditioners which consume a lot of energy. It is a lucrative and competitive activity,” he says, recalling the importance of quickly finding green technologies to reduce the industry’s dependence on fossil fuels.
On this side, John Plassard believes that the companies Schneider Electric (SU, 156.52 euros, Paris Stock Exchange) and Vertiv Holdings (VRT, US$37.50) should be considered, as should the Danish Asetek A/S (ASTK , 8.68 Norwegian crowns, Oslo Stock Exchange) and the Swedish Munters Group AB (MTRS, 139.00 Swedish crowns, Stockholm Stock Exchange).
EXPERT ADVICE
Rely on suppliers to avoid fluctuations linked to technological innovations
Mathieu Bouthillier, president and portfolio manager, B-Cap (Photo: courtesy)
“By investing indirectly in AI through component suppliers, we are not trying to predict who will have the best technology. It’s a bit like betting on a company that sells equipment to all the biotechs rather than betting on a company whose stock can jump 1000% if it finds the right molecule, but plunge 90% if the work searches are unsuccessful.
Titles to put on your radar
Advanced Micro Devices (AMD, US$102.25)
Alphabet (GOOGL, US$129.88)
Amazon (AMZN, US$133.26)
Applied Materials (AMAT, US$144.36)
Asetek A/S (ASTK, 8.68 Norwegian crowns, Oslo Stock Exchange)
ASML Holding (ASML, US$651.01)
Broadcom (AVGO, US$851.82)
Cadence Design Systems (CDNS, US$232.51)
Marvell Technology (MRVL, US$53.50)
Microsoft (MSFT, US$322.98)
Munters Group AB (MTRS, 139.00 Swedish crowns, Stockholm Stock Exchange)
Nvidia (NVDA, US$460.18)
Schneider Electric (SU, 156.52 euros, Paris Stock Exchange)
Synopsys (SNPS, US$442.24)
Taiwan Semiconductors Manufacturing Company (TSM, US$93.10)
Vertiv Holdings (VRT, US$37.50)
